PC Musician
Technique : PC MusicianThe ideal hard drive for music recording is quiet, reliable and very, very fast. Here's how to achieve it by choosing and setting up your system in the best way.
Martin Walker
A few years ago, achieving reasonable numbers of simultaneous tracks when recording and playing back audio files meant buying expensive SCSI drives, and getting involved in the extra hassle of installing and configuring a SCSI host adaptor card; and SCSI still has a place for the musician. It's an ideal way of transferring lots of data from one device to another via a cable of up to several metres in length, which makes it perfect for external drives, whether connected to samplers or computers (although I suspect FireWire drives are fast taking over in this arena). It's often an easier option if you want to use more than four drives simultaneously, since SCSI can run 15 or more from the single IRQ required by its host adaptor card. However, for the vast majority of internal PC hard drive duties, the once-humble EIDE hard drive is now quite fast enough to provide dozens of audio tracks, is still considerably cheaper, and can be far easier to use.
Choosing An EIDE Drive
However, if you do want to achieve high numbers of simultaneous tracks from an EIDE drive, there are a few things to watch out for. First of all, remember that musicians need high sustained transfer rates for continuous recording, and this factor is directly related to spindle speed. There seem to be plenty of 5400rpm drives available at the moment, but buying a 7200rpm drive is likely to give you faster sustained audio performance.
Don't be misled by statistics concerning huge peak transfer rates. The latest interface standard is ATA133 (often incorrectly referred to as UDMA133), and the first ATA133-compatible drive to be released was Maxtor's D740X. However, buying an ATA133 drive doesn't mean that you're likely to ever achieve 133Mb/second sustained transfer rates. It just means that if your motherboard, drive, and connecting cable all conform to this standard it's possible that when requesting data that's already in the hard drive buffer from a previous read, it can be shot across the buss at up to 133Mb/second. Audio recording requires sustained reads and writes that involve far more data than will fit in any hard drive buffer, so these peak rates are somewhat irrelevant. It was some time before ATA66 drives finally exceeded the ATA33 standard of 33Mb/second for sustained transfers, and in practice, I've yet to measure a sustained transfer rate on any drive that yet exceeds the sustained capability of UMDA66 -- typical recent drives like Seagate's Barracuda IV gave me DskBench readings of about 40Mb/second for both sustained read and write transfer rates.
Thankfully, all ATA standards are backwards-compatible: you can run an ATA133-capable drive in ATA100 or ATA66 systems and in most cases its sustained performance will be identical, so don't worry if your system doesn't support the latest mode. Indeed, I know of at least one specialist music retailer who deliberately runs some ATA100 audio drives in UDMA Mode 2 (ATA33), finding that this is the best way to get the maximum number of simultaneous recording and playback tracks.
Hard Drive Format Options
When you buy a new drive, it will already be physically pre-formatted into a number of tracks, each of which comprises a number of identically sized sectors. Because the outermost tracks have space for more sectors, more data can be read in a single revolution of the drive, and so the outer tracks always provide faster performance.
However, you will need to organise it to make best use of all its storage space. This process is called logical formatting, and combines a number of sectors into a 'cluster', and then maintains a File Allocation Table (FAT) pointing to the start of each file. FAT16 was the only file system option in Windows 95, and had no options for cluster size: this was determined solely by the size of the partition, up to a maximum of 2Gb, where cluster size became 32K. For a partition containing Windows and various applications, this could be extremely wasteful, since a one-byte file would still occupy a single 32K cluster.
A few diehard musicians still format their Windows drive as FAT16, since it has a marginally lower overhead due to a simpler file system, but it has largely been superseded by the newer FAT32 format first seen in the Windows 95B (ORS2) release. FAT32 overcomes the annoying 2Gb maximum size available to FAT16 partitions, but just as important for musicians, it lets you choose cluster size to make more sensible use of your storage. For a partition containing Windows and applications, which will probably contain loads of small files as well as large ones, using a smaller cluster size such as 4K will make more efficient use of space, while for partitions containing only data, where there are lots of tiny files such as graphics of saved web pages, you could use an even smaller cluster size -- I use 1K to make best use of space.
Where huge audio files are concerned, opting for larger 32K clusters will result in slightly less overhead when reading and writing files. However, I've never been convinced that it makes as big a difference as some people claim -- when I carried out some comparative tests way back in SOS December 1998, I measured an improvement of only about 1 percent in sustained transfer rate when increasing cluster size from 4K to 32K, and with today's faster drives the difference will be even smaller. Even so, it makes sense for most musicians to continue to format their audio drives using FAT32 with 32K cluster size to ensure low fragmentation, while using 4K for Windows, and between 4K and 1K for data partitions.
If you're running Windows NT, 2000, or the latest XP, you also get the option to format your drives using NTFS (New Technology File System), and even the option to convert them from FAT32 during installation. NTFS provides more security in systems that get used by various people, by keeping multiple copies of its master file table to protect against corruption and data loss. Although it's slightly more complex than FAT32, you are unlikely to notice much difference in performance, especially if you format using 32K clusters. The main difference is that while an NTFS partition is recognised by Windows NT, 2000 and XP, the Windows 95/98/ME platform won't see it at all. So don't use NTFS on data partitions if you have a multi-boot PC when one of your operating systems in Windows 98 SE or ME. Personally, I'm sticking with FAT32 for all my partitions, so they remain visible to all Windows operating systems. Remember that if you have PowerQuest's Partition Magic utility, you can convert from FAT or FAT32 to NTFS at any time without disturbing existing data.
The Silent Minority
The other feature most musicians find essential in hard drives is low acoustic noise. Although you can buy drives with spindle speeds of 10,000rpm or even faster (mostly SCSI ones), both noise and temperature tend to be directly related to spindle rotation speed. Even if the drive is quiet but hot, you'll have to cool it, and this means pulling cool air past it using fans -- another source of noise.
Thankfully, many drive manufacturers have now realised that home users like peace and quiet, and various EIDE models have appeared with special attention paid to lower noise and heat generation. Fujitsu's fluid-bearing models are very popular for this reason, although of late they have become increasingly hard to get hold of. Maxtor's D740X is a 7200rpm drive now available in a version with a Fluid Dynamic Bearing motor system, but unfortunately it gets quite hot, so you may end up generating additional noise cooling it down. The range that currently seems to hold the crown as quiet and available is Seagate's Barracuda IV, as used by many specialist music PC retailers including Millennium and Red Submarine.
There are also various measures you can take to minimise the noise contribution of any drive once it's inside your PC. As soon as you bolt a drive into your PC, its vibrations will get transmitted to the rest of the case, which can act like an amplifier. The obvious solution is to decouple the drive from the case using some sort of floating mount. There are commercial solutions, such as the NoVibes drive cradle I described in SOS January 2000, but the simplest and cheapest is to mount the drives on rubber or felt grommets instead of bolting it firmly to the drive bay. Exactly the same technique can be used with cooling fans, and you can find a detailed description of this DIY technique at www.7volts.com, along with loads of other useful information about keeping your computer silent.
One of the most annoying components of hard-drive noise is the seek noise of the read/write heads darting about when accessing the various chunks of audio data for each track in a song during playback. If this noise finds its way onto an acoustic recording it's next to impossible to remove using noise-reduction plug-ins, and perversely, defragmenting your hard drive can actually make it worse, as we'll see later on.
Some drive manufacturers provide a way to reduce seek noise: IBM's Deskstar 75GXP and more recent series, for instance, have what they term Automatic Acoustic Management (AAM), adjusted using a special DOS-based utility. This can result in a useful reduction of seek noise, but at the expense of a slight performance hit. In my experience, the two most successful ways to reduce both direct and structure-borne noise are either to start with a very quiet drive, or to mount each one inside a SilentDrive acoustic sleeve, available from various outlets including www.quietpc.com and used by music PC retailers including Carillon and Red Submarine.
Changing Channels
Once you've bolted your new drive into a suitable internal bay or SilentDrive sleeve, and attached the power cables, you'll have to decide which drive interface ribbon cable to plug into it, to connect it to your motherboard. In an ideal world, each hard drive, CD-ROM, and CD-R/W and DVD drive would have its own IDE channel and a separate cable, but although some PCs arrive with four IDE channels, most still only have two, labelled Primary and Secondary, each of which can support two devices configured as Master and Slave.
You can also buy IDE Controller cards supporting four extra IDE channels, such as the Promise Ultra 133 TX2, at about £60. These fit into a PCI expansion slot, and provide two more IDE connectors so that you can plug in up to four more IDE devices. This is one way to add Ultra100 or Ultra133 capability to an older PC, although since the real-world benefits will be minimal, and you'll also need another IRQ, it's only worth doing if you actually need extra IDE channels.
The boot drive containing Windows and your applications should always be connected as Primary Master, and if you are installing a second drive for audio purposes only, this should be connected as Secondary Master to keep it on a separate channel, since only one device on each channel can be accessed at any time (either the Master or the Slave). By separating your two hard drives onto different IDE channels, you will thus ensure that your audio data streaming is never interrupted if, for instance, the Windows swap file needs to be accessed.
If you run GigaStudio or HALion, you don't really want your streaming audio library on the same drive as your recorded audio tracks: this will degrade performance of them both, as the read/write heads are thrashing about from one part of the drive to the other. However, with two drives, the only alternative is to place your library on a dedicated partition on your Windows drive. Although not a perfect solution, this is what I've done as in practice, the Windows partition is unlikely to be accessed very often by your MIDI + Audio sequencer, as long as you have sufficient RAM installed to avoid the swap file being accessed.
I installed my CD-R/W drive as Primary Slave to keep it on a separate channel from the audio hard drive (where the data used to burn CDs usually comes from). Ignore warnings from people who tell you that this will slow down the transfer rate of your Primary Master drive to that of the CD, since this is a relic from the past, and hasn't applied for some years. Finally, if you have a separate CD-ROM drive, this can be connected as Secondary Slave. Placing it on the other IDE channel from the Windows boot drive should speed up large application installs from the CD-ROM, and if you copy CDs from the CD-ROM to the CD-R, it helps to have them on different channels.
In the nonstandard world of the PC, there are occasionally exceptions to these rules, and it pays to test your drives with DskBench to check that their performance is what you expected -- sometimes putting both hard drives on the same IDE channel can actually resolve speed problems.
Drive Cables
To successfully use any hard drive faster than ATA33, both your motherboard and BIOS have to support it, and you'll also need a suitable operating system that's capable of DMA transfers -- nearly all of the modern ones including Windows 98, ME, 2000 and XP do. Finally, you'll also need an 80-way ribbon cable connector, rather than the 40-way ones used by ATA33 and ATA66: the 80-way version uses the 40 extra paths as ground connections interleaved with the data lines, improving shielding, and thus ensuring data integrity for the faster rates. The motherboard recognises the difference by a break in one of the connections. You can also use 80-way cables for ATA33 drives, and this may possibly improve their performance very slightly for the same reason -- some drive manufacturers including Maxtor recommend this.
Either end of a 40-way cable can be plugged into the motherboard, and the Master device should be plugged in at the far end, while any Slave device should be attached to its middle connector. The 80-way version is a bit fussier: the blue connector must be plugged into the motherboard, the black connector at the other end used for a Master device, and the grey one in the middle for a Slave device. Bear in mind that to maintain data integrity and timing margin, the official maximum length for the 80-way cables is 18 inches, with the middle connector normally spaced 12 inches from the motherboard end. If you're using SilentDrive sleeves, or a nonstandard drive-mounting arrangement, this may not be long enough, and some retailers do sell 24-inch versions and even 36-inch ones -- but use the shortest one you can get away with. If you can manage with an even shorter length than the standard 18 inches, your hard-drive transfers may be marginally more reliable.
Check the manufacturer's details for how to set your drive for Master or Slave operation. This will nearly always be via jumpers on its back panel. Some drives have a 'Cable Select' option, in which case, as long as you are using 80-way cables, you can simply set both drives on the same cable to Cable Select mode and they should automatically be configured as Master and Slave depending on their positions on the cable.
New Mode
It's worth checking what mode your drive is set to when it arrives, as some manufacturers err on the side of caution and ship their ATA100 drives set to operate at ATA66 or even ATA33, to ensure they operate in all systems without errors. If your BIOS has been set to auto-detect your drives, you can determine their current UDMA mode settings by checking the BIOS readout as your PC boots up, and relate them to the various ATA standards by referring to the table, above right.
However, the most reliable way is to use the free utility provided by most drive manufacturers, which will also let you change the mode to suit your system. Some manufacturers bundle such utilities with their drives, but buying cheaper OEM rather than retail boxed models will probably mean that you have to download it yourself from the appropriate web site. You'll also need to transfer the files to a floppy disk, and use this to boot up your PC when switching it on for the first time after physically installing the drive, since it's not possible to change UDMA mode from within Windows. If you change it, you'll have to power down your system and then restart it -- pressing the reset button isn't enough to let the new mode 'take'.
Don't assume that if you just use the drive 'as is', and it works, that you don't need to check its UDMA mode. I once installed an ATA66 drive in an ATA33-only PC, and it seemed to work well, but DskBench soon proved that its transfer rate was far slower than expected, and dropping it from ATA66 to ATA33 gave it a much faster sustained transfer rate. As mentioned earlier, running some drives as ATA33 may give them better full-duplex performance when both recording and playing back lots of simultaneous tracks than even ATA100.
By the way, don't forget to revise the optimum ATA mode for your drives if you move them to
RAID Arrays
Sooner or later, most musicians with more than one hard drive will wonder about the possibility of sharing audio recording and playback duties between them, to achieve more simultaneous tracks. This is of course the way office networks operate, with banks of hard drives, but in this scenario it's for rather different reasons. First, they need to store a vast amount more data than any home or studio computer, and therefore must split it across many drives. Second, it's vital that the network can carry on working even if one of the drives ever fails, since so many people may depend on it.
The answer is not just to split the data across drives, but to duplicate some of it, so that even if one drive breaks down, the remainder still contain 100 percent of the data between them -- this is termed redundancy. These two factors are what make RAID (Redundant Arrays of Inexpensive Disks) so popular in the business world.
However, storing redundant data gives a performance hit, and would rarely provide any advantage during audio work, except possibly in the case of a once-in-a-lifetime live performance where you want the extra security. Also, it's highly unlikely that you would need more storage space for a single project than that provided by a single modern hard drive. Of course, you can use additional drives for backup purposes, but even a 10Gb drive can hold about an hour's worth of continuous 32-track 16-bit/44.1kHz recording, while one of the new 80Gb drives could hold about 2.5 hours' worth of continuous 32-track 24-bit/96kHz recording.
The only remaining advantage of using RAID for audio might be sharing the workload, but with many modern drives achieving 40Mb/second sustained transfer rates, which typically equates to more than 50 simultaneous tracks of 24-bit/96kHz recording when using a 128K buffer, it just isn't worth the extra hassle unless you really need cutting-edge performance.
UDMA Modes & Burst Data Transfer Rates
UDMA Mode 0: ATA16 (16.7Mb/second).
UDMA Mode 1: ATA25 (25Mb/second).
UDMA Mode 2: ATA33 (33Mb/second).
UDMA Mode 3: ATA48 (48Mb/second).
UDMA Mode 4: ATA66 (66Mb/second).
UDMA Mode 5: ATA100 (100Mb/second).
UDMA Mode 6: ATA133 (133MB/second).
a new PC -- while researching this feature I realised that my 30Gb Seagate drive was still set to UDMA Mode 2 (ATA33), although in my new PC it's perfectly capable of being run in Mode 3 (ATA66).
Buss Mastering & Performance
Before you do any serious work with your new drive, check that DMA Buss Mastering has been enabled, since your CPU overhead while accessing the drive will plummet from perhaps 50 percent to less than 1 percent. Buss Master support needs to be enabled for each EIDE drive (and any non-SCSI CD-ROM drives); in Windows 98 you do this from the Disk Drives section inside Device Manager. Click on the Properties button for each EIDE drive, select the Settings tab, and then make sure that the DMA box is ticked. In XP the section is labelled IDE ATA/ATAPI controllers. Right-click on each IDE channel entry, and select the 'Advanced Settings' page -- here you can check that the Transfer Mode is correctly set to 'DMA if available'.
Some hard drive controllers, especially those on PCI expansion cards, enable DMA Buss Mastering automatically, but it's worth checking anyway just to make sure. As always, the best utility for the job is DskBench, which simulates the particular environment of audio recording and playback by creating a set of large files which are read in sequential chunks, just like audio tracks in a sequencer, as well as providing a readout of CPU overhead.
Typical ballpark figures for the sustained transfer rate for a 5400rpm drive are perhaps 20Mb to 30Mb per second, and 30Mb to 40Mb per second for one spinning at 7200rpm, all with CPU overhead of well under 5 percent if DMA Buss Mastering has been correctly enabled. As you can see, even the fastest of these only needs an ATA66-capable system. In theory, a 40Mb/second drive can manage about 160 simultaneous 16-bit/44.1kHz audio tracks with a buffer size of 128K, and about 50 at 24-bit/96kHz, although in practice this will depend on other factors including any fragmentation of the drive contents.
Frag Factors
Most musicians understand the need to periodically defragment their hard drives. As files get deleted, they leave gaps which then get filled when another file is saved. However, once the newer file exceeds the size of the gap, the remainder is saved into the next gap found, and so on, so that an individual file may end up split into fragments scattered across the drive. When you load this file, it will take considerably longer than it could, simply because the drive read/write heads are jumping about all over the place to access all its segments. By defragmenting the entire contents of the drive so that each file is contiguous (ie. in one block), you speed up loading times.
Some utility programs, notably the Norton Utilities suite, go somewhat further, by simultaneously rearranging the files according to user options. Frequently accessed files can be placed at the outside of the drive where its performance is speediest, while infrequently accessed files can be placed at the slowest inside portion of the drive, leaving whatever free space remains between the two. This technique was taken further by Microsoft in Windows 98 and ME with the Application Launch Accelerator. Accessed from the 'Rearrange my program files so my programs start faster' option in the OS defragmentation utility, this rearranged application files not so that they were all individually defragmented, but so they were arranged contiguously in the order that the files were required when the application launched. It did this by monitoring application loads and then rearranging files, perhaps even in smaller chunks that caused some slight defragmentation, but so that you noticed an improvement in progam launch times.
You can see the optimisation files for each application in the Windows 98/ME Windows / Applog folder, and view the optimisation log file by opening Optlog.txt to see how the various applications have been reordered. Periodically, you can delete the entire contents of this folder to force Windows to start re-optimising afresh and ignore older applications that may since have been deleted. One disadvantage of this approach was that since many system files are used by numerous different programs, start times of some could be optimised at the expense of others.
Windows 2000 didn't include this technology, and Windows XP once again returns to the more straightforward defragmentation approach. However, it adds a boot optimise function, which places all the files required for booting your particular system next to each other on the drive, to speed up the process. It will start optimising after the first boot, so your second boot should be faster. However, while Windows XP is constantly fine-tuning file positions on each boot, Microsoft estimate that 90 percent of the optimisation is done in the first two.
Swap File Options
While rearranging the contents of your hard drive, a defragmenter can perform various other tasks that further improve disk performance. One of the best for musicians who have decided to create a permanent swap file (or paging file in Windows NT, 2000 and XP) of a fixed size is to move this to the fastest outside area of the drive. Being fixed in size, it can never become fragmented either.
Some defraggers, like Norton Utilities' Speed Disk, provide this option, although sadly the various defraggers built in to different versions of Windows don't. However, many musicians including me are beginning to question the effectiveness of running Norton Utilities on their music partition, since utilities as Live Update and Live Advisor are redundant in the stripped-down environment of a music PC. In addition, many of its features default to running in the background, and although this behaviour can be defeated, you do end up with the feeling that you're disabling as much as you're using.
One solution for those who are running Norton Utilities on their main Windows partition, but have created a multi-booting system with a separate, leaner music partition, is to run Speed Disk 'across the divide' by forcing your music partition to be visible from your main one. You can do this with a utility like Partition Magic (with its Unhide Partition function), but if like me you're also using the associated BootMagic utility to switch between partitions, you'll have to first temporarily disable this by unticking the BootMagic Enabled box using its Configuration utility. Otherwise, as soon as you reboot and choose the main partition, it will re-hide the music one.
Even so, this method has limitations. According to my tests, even if you set up a permanent fixed-size swap file on your music partition, Speed Disk doesn't recognise it when running from another partition, and treats it as just another file. However, there is still a way to force the swap file to be moved to the outside of the partition. In Speed Disk's Options page, click on the Customize button in the Optimisation Method area, click on the Files First page, and browse to the file named WIN386.SWP in the root folder of your music partition (or PAGEFILE.SYS if running Windows NT, 2000, or XP), then click on the OK button.
I've never personally agreed with any of the web sites that recommend you to arbitrarily set up a swap file of 1.5 times the size of your RAM, especially now so many of us have such large amounts installed. As I explained way back in SOS January 1999, there's a much more scientific way to find out the optimum swap-file size for your PC. Indeed, if you already have 512Mb or more of RAM, opinions are somewhat divided about the benefits of having a swap file at all, since all your applications should have enough space to run in RAM simultaneously without using one. Be warned, though, that if Windows ever exceeds your RAM quota without a swap file available, you're likely to have a bad crash. So, if you're cautious, you could just set swap file-size to a nominal 50Mb or so, but if you want to go for it, just select the Disable Virtual Memory option in Windows 98/ME, or the No Paging File option in Windows 2000/XP.
Audio Defragmentation
Another reason I've written a recap on defragmentation is that for best music performance, audio files should ideally be defragmented in a different way. The excellent DskBench utility benchmarks hard drive performance by creating a set of eight large files and then times reading them a block at a time, exactly as audio applications do. As the block size decreases, so the read/write heads have to jump about more often between blocks, and so the maximum number of simultaneous audio tracks also decreases for a given hard drive. This is why choosing a suitably large block buffer size in your audio application will give you more audio tracks, but at the expense of using more RAM.
However, once you 'defrag' a drive containing audio files into single long contiguous chunks, you may actually reduce performance when reading multiple tracks, since as each track is read a block at a time, the read/write heads may have to jump further to the appropriate block of the next track. The answer would be to have a special defragmentation utility that rearranged audio files into chunks whose size matches whatever audio block size you have chosen in your audio application.
For some reason none of the big music software developers have to my knowledge created such a utility. However, as often proves to be the case, an enterprising shareware developer rose to the challenge, and I recently discovered Interleave from AnalogX (www.analogx.com), a freeware utility that does exactly this if you modify its 'content size' from the suggested low settings to that of the Disk Block Buffer Size (see the screenshot). I tested the theory out by creating a new version of a song's Audio folder, and then creating a new set of specially interleaved WAV files in it. Sure enough, I noticed an audible reduction in head seek activity when playing back the interleaved version, which suggests that the drive wasn't having to work quite so hard. So, you may indeed manage a larger number of simultaneous tracks after rewriting your audio files in a carefully fragmented format.
However, the fact remains that for optimum hard drive performance, the audio data would be best reorganised by the sequencer, since only this knows the precise order in which the tracks are read, and of course the user's choice of block buffer size. So, Cakewalk, Emagic or Steinberg for example could incorporate an 'Optimise Audio Files for Current Song' function into their sequencers, which rearranged all the files so that the various audio sections were carefully laid out chunk by chunk to minimise head read/write activity, and therefore increase the number of available tracks. What about it chaps?
No comments:
Post a Comment